[Taxacom] The difference

[Pierre Deleporte]

Sorry I could'nt resist a few comments

- "total evidence" properly is not directly at stake; the problem is  
rather simultaneous analysis versus separate analyses; but separate  
analyses can (and should) use total relevant evidence anyway (a  
logical requirement for induction)

- redundancy between molecular and morphological data is certain only  
when one has the complete relevant genomic data set (or rather,  
"information")... merely complete genome sequencing? I'm not even sure  
of that, but not smart enough in genetics of the development to  
comment further... just heard that the genome is not alaways  
indifferent to cytoplasmic composition an that development processes  
can be tricky things...

- of course combining even partial molecular and morphological data  
sets leaves one at risk of uncontrolled redundancy... but also "at  
risk" of gathering complementary information, which is rather  
appealing and nicely fitting the "total evidence" requirement

- the risk of redundancy may be annoying, but redundancy itself is not  
necessarily misleading... why should random sorting obligatorily  
provide misleading signal, rather than informative one, not to mention  
some balanced neutralization? Has Richard Zander any probabilistic  
hint on this line? Evaluating risk of redundancy, but also risk of  
misleading, neutral or informative redundancy? Too much good should  
not hurt...

- weighting is a Q question, even inside molecular and morphological  
data sets themselves; but separate analyses will not help: how will  
you "balance" discordant topologies for a synthesis? This is called  
weighting, I bet... And if you can adjust your weighting for comparing  
separate analyses, everything is ready for a combined analysis, which  
is simply the smartest way to balance separate analyses of complex and  
partially discordant data sets: just do it in one click

- now I agree with Richard and Richard on two points: 1) weighting  
morphological traits against molecular ones is a serious question, and  
2) we generally don't know much yet about evo-devo processes to  
support even a grossly tuned weighting... beyond some notion that  
morphological characters should likely "carry much", but I bet that  
some tiny changes in development regulation genes can also have  
tremendous phenotypic consequences for some characters... not to  
mention homoplastic evolution this way...

- the important point in my view is that we go on trying and thinking  
in such terms: neither "equal weights and combine because we don't  
know" (why not random weights and random partitioning, if we really  
don't know anything about anything?), nor "combining is forbidden"  
(sorry, it's really nonsense to me, it all depends on what you know or  
can bet...); but: let's try and know better about probabilities of  
changes according to general evolutionary rules (if any...) and  
documented redundancy, and implement what we pretend to know. This  
should raise some hope of overcoming a simplistic "optimality alone"  
option, which I consider to be "equal weights for want of knowing  
better".

- fortunately I don't think everybody stopped puzzling on  
morphological, or even behavioral cladograms, despite some fashionable  
despising.

thanks fo stimulating thread
Pierre


Quoting Don.Colless at csiro.au:

> Although not very active in the field any longer, I must enter my   
> long-held opinion that mixing molecular and morphological data sets   
> is quite illogical - on the simple grounds of redundancy and   
> probable grossly unequal weighting of characters. If the genomic   
> composition is ultimately responsible for total morphology (the   
> Hennigian holomorph), but with the fine detail quite unknown, then   
> said mixing simply entails a complex  network of double entry   
> without weighting. So, why not stick to parallel studies of the two,  
>  with subsequent analysis of any serious differences. This business   
> about "total evidence" is nonsense if it entails the effects I've   
> mentioned.
>
>
> Don Colless,
> Div of Entomology, CSIRO,
> GPO Box 1700,
> Canberra. 2601.
> Email: don.colless at csiro.au
> Tuz li munz est miens envirun
>
>
> 	-----Original Message-----
> 	From: taxacom-bounces at mailman.nhm.ku.edu on behalf of Richard Zander
> 	Sent: Sun 28-Oct-07 7:48 AM
> 	To: Richard Pyle; taxacom at mailman.nhm.ku.edu
> 	Cc:
> 	Subject: Re: [Taxacom] The difference
>
>
>
> 	Rich:  Thanks for the follow up. I do think that the author of   
> every paper that analyzes a combined morphology and molecular data   
> set does make a decision on the number of apomorphic shared DNA   
> bases implied by a shared apomorphic morphological trait, namely   
> one. This could be wrong by two orders of magnitude. Even if a   
> shared morphological trait were recognized as a surrogate for,   
> say,only  10 shared molecular traits, the fabric of modern   
> phylogenetic classification would necessarily change.
>
> 	1. Combined data sets would suddenly not swamp the morphological component,
> 	and if the data sets were not combined, then
> 	2. morphological data sets would suddenly have high bootstraps and   
> maybe statistically certain BPPs on branch arrangements.
>
> 	(Remember when "optimality alone" of morphological data sets was in  
>  vogue as justification for using cladograms in classification?  
> Maybe  we are back to puzzling over morphological cladograms.)
>
> 	________________________________
>
>
> _______________________________________________
> Taxacom mailing list
> Taxacom at mailman.nhm.ku.edu
> http://mailman.nhm.ku.edu/mailman/listinfo/taxacom
>



-- 
Pierre Deleporte
UMR 6552 EVE
Station Biologique
35380 Paimpont
tél 02 99 61 81 63
fax 02 99 61 81 88